[lsoda] error

[FanWn]

@kylebaron Hi Kyle, I'm trying to optimize two parameters of my PBPK model against some observed data (11 mean concentration data points). The codes are as follow:

data<-read.csv("data.csv")

yobs <- filter(data,EVID==0) %>% dplyr::select(DV) %>% unlist %>% unname
yobs

wss <- function(dv, pred, par=NULL) {
  sum(((dv-pred)/dv)^2)
}

pred <- function(.par, .data, yobs=NULL, pred=FALSE) {

  .par <- .par %>% setNames(names(theta))

  .mod <- param(mod,.par)

  if(pred) {
    out <- mrgsim(.mod,data=.data,carry.out="type", maxsteps = 10000)
    return(as_data_frame(out))
  }

  out <- mrgsim(.mod, data=.data, obsonly=TRUE, Req="Cvenous", maxsteps = 10000)

  return(wss(yobs,out$Cvenous))

  #return(-1*sum(dnorm(log(yobs),log(out$CP),.par$sigma,log=TRUE)))

}

theta<-c(CL_r=9.8, Kp=0.001)
fit1b <- optim(theta, pred,.data=data, yobs=yobs)

Then I got the lsoda error:

[lsoda] warning..internal t = 1.25203 and h = 6.11389e-17 are such that in the machine, t + h = t on the next step solver will continue anyway. [lsoda] warning..internal t = 1.25203 and h = 6.11389e-17 are such that in the machine, t + h = t on the next step solver will continue anyway. [lsoda] above warning has been issued 2 times it will not be issued again for this problem. [lsoda] 10000 steps taken before reaching tout; consider increasing maxsteps.

[mrgsolve] lsoda returned with negative istate: -1 excess work done on this call; check the model or increase maxsteps. current value of maxsteps: 10000

Error: simulation terminated.

I tried to increase the maxsteps value but that didn't work and the same error occurred. Can you tell what is wrong? Is it because there is not enough observed data?

[kylebaron]

Hi @FanWn -

warning..internal t = 1.25203 and h_ = 6.11389e-17

It looks like the ode solver is having an issue at time = 1.25 so I don't think it should take that many steps to get there.

The first thing I'd try doing is log-transforming the parameter list

data<-read.csv("data.csv")

yobs <- filter(data,EVID==0) %>% dplyr::select(DV) %>% unlist %>% unname
yobs

wss <- function(dv, pred, par=NULL) {
  sum(((dv-pred)/dv)^2)
}

pred <- function(.par, .data, yobs=NULL, pred=FALSE) {

  .par <- lapply(.par, exp)

  .par <- .par %>% setNames(names(theta))

  .mod <- param(mod,.par)

  if(pred) {
    out <- mrgsim(.mod,data=.data,carry.out="type", maxsteps = 10000)
    return(as_data_frame(out))
  }

  out <- mrgsim(.mod, data=.data, obsonly=TRUE, Req="Cvenous", maxsteps = 10000)

  return(wss(yobs,out$Cvenous))

  #return(-1*sum(dnorm(log(yobs),log(out$CP),.par$sigma,log=TRUE)))

}

theta<- log(c(CL_r=9.8, Kp=0.001))
fit1b <- optim(theta, pred,.data=data, yobs=yobs)

Can you try that and let's go from there.

Kyle

[FanWn]

Hi Kyle @kylebaron , thanks for troubleshooting! The codes worked with what you suggested (only the function for pred should be pred <- function(par, .data, yobs=NULL, pred=FALSE) {). However, the simulation is still off after I plugged in the optimized parameter values. I tried a couple of different initials but didn't improve.

[kylebaron]

Ok; thanks for reporting back. Next, I'd look to see what might be going on in the data or model; this will need to be shared so we can take a look.